xmodar/commonvoice-12.0-arabic-voice-converted

该数据集是从Common Voice Arabic Corpus 12.0派生而来的，包含了自动添加音标的转录和音素表示。数据集中的录音是由用户朗读的阿拉伯语文本，这些文本最初没有音标，可能导致阅读错误。音标和音素是自动生成的，因此数据集对于语音识别任务有价值，但存在固有的噪声。数据集创建于SDAIA冬季学校，适用于对现代标准阿拉伯语的音标化语音数据和语音转换音频感兴趣的研究人员和从业者。音频文件最初没有说话者ID，因此使用了xTTS-v2模型提取说话者嵌入，并进行了聚类和语音转换。数据集中的录音是由志愿者贡献的，没有新增录音，仅包含现有文件的处理版本。数据集包含来自阿拉伯世界不同方言的录音，但缺乏具体的方言或人口统计信息，音频质量较差，可能存在丢段、背景噪音、音调扰动、阅读错误和自动生成的音标等问题。

This dataset is derived from the Common Voice Arabic Corpus 12.0 and includes automatically diacritized transcriptions and phoneme representations for augmented audio data. The recordings feature Arabic text read aloud by users, where the text was initially undiacritized, allowing for potential reading errors. The diacritization and phonemes were generated automatically, resulting in a dataset that is valuable for speech recognition tasks but inherently noisy. The dataset was created by adapting and performing voice conversion on the dataset provided by @mostafaashahin as part of the SDAIA Winter School held at King Saud University, Riyadh, in December 2024. The audio files lacked speaker IDs, so speaker embeddings were extracted using the voice_conversion_models/multilingual/vctk/freevc24 model from xTTS-v2. These embeddings were clustered and then used for voice conversion, enhancing the dataset for further research in speech processing. The original recordings were contributed by volunteers as part of the Common Voice Arabic Corpus 12.0. The dataset includes recordings from various dialects across the Arab world, but specific demographic or dialectal statistics are not available. The audio quality is suboptimal, with issues such as dropped segments, noisy backgrounds, perturbed pitch, potential reading errors, and automatically generated diacritization, which may impact certain tasks requiring high-quality, clean data.